GEO Casa · deck 2 of 4
GEO by role
Boss · marketing · technical lead · editors · reporting · PR
Use ← → or swipe to move between slides
Boss: what GEO actually is
Three things to know before you approve a budget.
Google: for Google Search, optimising for generative AI search is optimising for the search experience, "and thus still SEO".
Official · Google Search CentralGoogle: AI Overviews and AI Mode have no extra technical requirements. A page needs to be indexed and eligible to show with a snippet.
Official · Google Search CentralEven a vendor-written ROI white paper says GEO should not promise fixed rankings, a fixed recommendation order or fixed sales.
Hypothesis · 一搜百应 GEO 研究院
Pay for effort and measurable change. Rankings can't be bought.
Boss: the promise to refuse
A good agency can guarantee we are the first recommendation.
Answers change from run to run. Even the provider in this case found that a "stable first place" did not match what it observed. Ask for measured movement on a fixed question set instead.
A provider tracking one client studio saw its recommendation rate move between 10% and 68% within two months, ending at 40%.Observation · 2026-08 · 合肥图灵象人工智能科技有限公司
Boss: what to ask, what to refuse
Three questions for your team or agency, and three promises nobody should make.
- Which questions and platforms are we measuring?
- How many runs per question, and what range?
- What do Search Console and Bing report?
- Refuse: a guaranteed recommendation on any platform
- Refuse: a set amount of sales or leads in the short term
- Refuse: single-run screenshots as proof
Only accept results you can re-test.Read the full guide
Whoever signs the agency contract
Buying from an agency: the contract
- 1Fix the platforms, question set and competitors up front
- 2Write metric definitions, sample size and re-test rules into acceptance
- 3Keep traffic, leads and sales out of the provider's promises
- 4Ban fake cases, reviews and endorsements in writing
- 5Own the data; agree exit and handover
- 6Accept checks by independent monitoring
Acceptance means re-testing on the agreed question set.Read the full guide
Buying from an agency: the red lines
NBD's GEO Red Book: GEO should not promise a fixed recommendation on any platform, set short-term sales or leads, or single-run screenshots as proof.
Observation · 2026-06 · 每经AI智库CCTV's 3·15 show: a GEO tool published 10+ articles about a fitness band that did not exist, and two major AI models then recommended it near the top.
Observation · 2026-03 · 第一财经Google's definition of spam includes "attempting to manipulate generative AI responses in Google Search."
Official · Google Search Central
If a tactic would embarrass you in the news, don't let the agency use it.
Marketing: a 90-day plan from zero
The first 90 days buy a map and a clean foundation. Mentions come later.
Read the full guideMarketing and brand managers
Marketing: what the 90 days contain
- 1Write down what AI should say, and 20–30 real questions
- 2Baseline across the assistants customers use
- 3Confirm access to your key pages
- 4Rework three to five core pages
- 5Get listed correctly on cited list sites
- 6Re-measure the same way and report foundations
Measure the same questions the same way every round.Read the full guide
Who does what, month by month
The same 90 days, split by role. A suggested split, not a measured one.
- Month 1Set up
- Marketing: positioning and 20–30 questions
- Reporting: baseline on the prompt panel
- Tech lead: check crawler and CDN access
- Month 2Map and fix
- Reporting: map the sources answers cite
- Tech lead: fix rendering and snippet issues
- Editors: plan the core-page rework
- Month 3Build
- Editors: rework three core pages
- PR: get onto the top cited lists
- PR: one asset third parties can use
- Day 90 onReview
- Reporting: re-measure the same way
- Boss: judge foundations, not mentions
- Boss: ROI review around day 180
At day 90, report the groundwork done. Mentions usually come later.Read the full guide
Technical lead: can AI reach your pages?
Five yes/no checks on your most important pages.
1Does robots.txt allow Googlebot, Bingbot and OAI-SearchBot?
Allow search crawlers. Blocking training bots is a separate choice; OAI-SearchBot and GPTBot are set independently.
2Does the CDN or firewall let those crawlers through?
Check CDN bot settings, such as Cloudflare's AI-bot blocking. Google says crawling must be allowed there too.
3Is the key content in the HTML without running JavaScript?
Use server-side or pre-rendering. Major AI crawlers tested did not run JavaScript.
4Are key pages free of nosnippet and tight max-snippet?
Remove nosnippet from pages you want quoted. It keeps content out of AI Overviews and AI Mode.
5Are the pages indexed in Google and Bing?
Submit sitemaps and check indexing in Search Console and Bing Webmaster Tools.
If all five pass, crawlers can reach the pages. Content and earned coverage decide the rest.
Technical lead: search bots and training bots
OpenAI: OAI-SearchBot and GPTBot settings are independent. A site can allow OAI-SearchBot for ChatGPT search while disallowing GPTBot.
Official · OpenAIAnthropic separates ClaudeBot (training), Claude-User and Claude-SearchBot. Blocking the latter two may reduce visibility in Claude.
Official · AnthropicAn August 2026 audit of robots.txt on 130 company sites called blocking search and training crawlers together the most common, most damaging mistake.
Observation · 2026-08 · DATA17 AI Research Lab · 解数智能
Decide on training separately. Never block search by accident.
Editors: write passages that survive being quoted
Assistants quote passages from a page. Switch between the two versions.
Our innovative solution has helped many customers significantly improve their results over time. Learn more about our pricing below.
Studio A's beginner course costs ¥199 for four classes (prices checked October 2026). Classes run Monday to Friday until 9 pm, with at most eight people per class.
- Answer first
- Numbers with conditions
- Names, not "we"
- Dated
Illustrative example with a made-up business.
Editors: what the evidence says
In the 2023 GEO paper's benchmark, citing sources, adding quotations and statistics improved visibility by 30–40% (41% at best) on the authors' word-count measure.
Research · Aggarwal et al., arXiv:2311.09735 (KDD 2024)Microsoft: Q&A pairs can often be lifted "word for word" into AI answers.
Official · Microsoft Advertising Blog
Formatting makes good content easier to quote. It can't fix weak content.
Editors: E-E-A-T without the folklore
Four things editors often hear. Click each card for what Google says.
Google: of experience, expertise, authoritativeness and trustworthiness, "trust is most important."Official · Google Search Central
Editors: can we write with AI?
Google penalises content written with AI.
Google judges what content does for readers, however it is produced. The red line is generating many pages mainly to manipulate rankings, with no added value. Check every fact: models predict words and don't look facts up.
Google: "Appropriate use of AI or automation is not against our guidelines." Using it mainly to manipulate rankings is.Official · Google Search Central Blog
Editors: the scaled-content red line
Google: "Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users."
Official · Google Search CentralOn May 15, 2026 Google clarified that its spam policies "apply to all of Google Search, including generative AI responses."
Official · Google Search CentralA separate page for every way people might search, fan-out queries included, made mainly to sway rankings or AI answers, violates this policy.
Official · Google Search Central
A few pages people use are safer than a hundred thin ones.
Reporting: why one run proves nothing
Ask the same question again and the answer changes. A prompt panel asks each fixed question many times and reports the rate.
No answers yet.
Simulated answers for illustration, not real data.
Miaozhen (2,025 runs, three Chinese models) found results reach basic stability only at 30 or more runs per prompt; low-visibility brands need 60+.Observation · 2025-09 · 中国商务广告协会AI营销应用工作委员会 · 秒针营销科学院
Reporting: show the trend, the range and the sample
A mention rate from repeated sampling, with its uncertainty band.
Microsoft: Bing's AI Performance report "does not measure rankings, authority, performance, or importance."Official · Microsoft Bing
Reporting: the two official reports
First-party data from Google and Microsoft. Free, and the only view from inside.
Search Console's Generative AI performance report shows impressions of your links in AI Overviews and AI Mode, by page, country, date or device.
Official · Google Search Console HelpGoogle: sites in AI features "are included in the overall search traffic in Search Console", under the Web search type.
Official · Google Search CentralMicrosoft: trend changes in Bing's AI Performance report cannot be traced to a specific cause, update, model change or content edit.
Official · Microsoft Bing
Start every report with the platforms' own numbers.
Reporting: what GEO tools can see
A GEO tool can show our ranking inside Google's AI answers.
Tools can run your prompt panel and record what answers say. They cannot see inside Google. For Google, the inside view is Search Console's Generative AI performance report.
Google: "No third-party tool has access to our internal ranking or AI systems."Official · Google Search Central
Reporting: what goes on the slide to your boss
Promise effort and measured movement, never rankings.
- Search Console and Bing reports first
- AI referral traffic, reported as a floor
- A fixed prompt panel with enough repeats
- Leads asked "how did you find us?"
- One headline trend with its range
- Two or three before-and-after answers
Put the sample size next to every number.Read the full guide
PR: answers lean on what others publish
On "best cola" questions, major brands took 56.3% of ChatGPT's mentions and niche brands 12.3%.
Research · Chen, Wang, Chen and Koudas (University of Toronto), arXiv:2509.08919DataStory found cited content in nine consumer categories was mostly ranking and recommendation lists (22.8%–72.0%).
Observation · 2026-08 · 数说故事
A test: if the mention still stands once people know who arranged it, it was earned.
PR and brand teams
PR: which mentions do AI answers use?
- 1Map which sites answers in your category cite
- 2Earn coverage with real news, data and access
- 3Help reviewers describe you accurately; never buy rank
- 4Treat encyclopedias as rule-bound: qualify, disclose, cite
- 5Keep one set of facts everywhere
AI can only cite coverage that already exists.Read the full guide
PR: encyclopedias and reviews have rules
Wikipedia presumes a company notable after significant coverage in multiple reliable, independent secondary sources. Press releases don't count.
Official · WikipediaBaidu Baike restricts company sites, forums, blogs and self-media as references. Honours, funding and certifications need media reports or registry records.
Official · 百度百科The US FTC rule bans writing, creating or selling fake reviews, and covers buying reviews, insider reviews and fake social media indicators.
Official · US Federal Trade Commission (eCFR)
Meet the notability bar and disclose your interest. Self-written entries tend to be removed.
PR: seeding mentions
Seeding mentions across forums and blogs will get AI to recommend us.
Google says its core ranking systems focus on quality content and other systems block spam, and its AI features depend on both. In China, the 2026 Qinglang campaign targets GEO used for malicious marketing.
Google: "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem."Official · Google Search Central
Who owns what
One owner per task, with the role that supports it. The basis column grades the guidance behind each task.
| Task | Owner | Supports | Basis |
|---|---|---|---|
| Crawler and CDN access | Technical lead | Marketing | Official |
| Core pages that answer clearly | Content editors | Marketing | Official |
| Official AI reports | Reporting | Technical lead | Official |
| Prompt panel, 30+ runs per question | Reporting | Marketing | Observation |
| Earned coverage and cited lists | PR | Marketing | Research |
| Agency contract and red lines | Boss | Reporting | Observation |
| The 90-day plan | Marketing | Boss | Hypothesis |
The split of roles is a suggestion. The basis grades the guidance for each task, not the split.
How do the roles stay aligned?
Everyone measures with the same question set. Each slide condenses a full guide. Read the guide for the evidence behind it.